Training DPR with different language

[shahad2099]

Hello everyone
I want to train the DPR in a different language [Arabic]
I've been trying for almost two weeks but I feel like I'm lost :(
I want to know what are the steps and what should I change to train.
I know that I should change the bi-encoder into an Arabic model [Arabic version of Bert]
and also change the format of the datasets into DPR format.
other than that I feel lost what should I do next? should I do the next steps :
1- run generate_dense_embeddings.py with Arabic wiki corpus
2- then train train_dense_encoder.py with the Arabic Language model and Arabic dataset.
3-and finally, evaluate using dense_retriever.py.
Is that what I have to do?

Thank you

[zlz-123]

Hello, I also want to try using different languages to train DPR, and we can communicate with each other